While a lot has been covered with regard to what’s going on with DeepSeek, I feel it’s important to explore where we go from this point. Look at how visionaries reacted. Marc Andreessen called the release of the DeepSeek R1 chatbot “a Sputnik moment for AI”. Mark Zuckerberg convened 4 war rooms of Meta engineers to catch up with DeepSeek’s efficiency. US President Donald Trump called it a wake-up call for the US technology industry.
All of it made for great news and headlines. But I think that one can’t discount markets and take them for granted. $100 billion in a stock market sell-off. That kind of number can tangibly move the gross domestic products (GDPs) of some countries. Usually, markets never react so strongly to one technology startup competing with others. But, somewhere, what happened with Nvidia seemed to remind the markets of the Cisco moment of the dot-com bubble.
Also Read: OpenAI mulling massive $500 million deal with former Apple design chief to bring iPhone replacement
Cisco was the apple of one’s eye, contributing to the rise of the Internet as a phenomenon, but when the bubble burst, the conversation changed from undersupply to oversupply, leading to Cisco having to sell its inventory. What may happen with Nvidia is that there could be less demand for its newer and more expensive GPUs. There might be an oversupply of GPUs in 2025 akin to the oversupply of networking equipment circa 2001. The silver lining, however, was that it was great news for companies, like Facebook, Apple, Netflix, Google and others, which emerged from the ashes of the dot-com bubble.
With what’s going on, are we going to see AI 2.0 with a different style of AI startups?
Keith: What DeepSeek did really well was solve engineering problems. The US was stopping chips from going to China, so it was a case of “necessity is the mother of invention”, forcing China to rethink how to approach the problem. To some degree, it’s not a completely new AI, it’s a different way of doing things. I thought Nvidia was oversold and there’s still a lot of demand for its GPUs out there, so going forward, the Nvidia piece will be a good play. The tech bros – for lack of a better word – on the West Coast have been buying up all the chips, so they might end up looking a bit silly, because a lot of their valuations are based on the fact that they’ve created that beachhead and they’re assuming that that kind of chip capacity is a barrier to entry to others. In my case, I’ve added to my Nvidia holdings, while not doing that for any other tech stocks, so we’ll see how that plays out down the line. I was in the VC-PE world back in the 90s, so I do have the scars from the dot-com crash, but I think this is just a bump on the road and there’s a long way to go.
Bhandari: The DeepSeek team figured out an extremely efficient way to get things done. It may be imperfect or not exactly what was envisioned, but those are nuances that can be tweaked. The markets are reacting, because this technology has been getting mainstream. Having seen a few industry cycles, I think this is the consumerisation of the Internet. Way back when, the Internet was a technology used by enterprises and universities to exchange information and facilitate transactions. The onset of browsers and other elements made it a consumer product. Today, most users of AI platforms, like ChatGPT, Perplexity and Gemini, are using it for simple tasks, but there’s going to be a thermonuclear explosion in terms of usage, ideas and applications.
Also Read: ChatGPT Plus available to these users for free, here’s how to enroll
So, this market reaction is largely centered around a company or two deemed to be invincible and valued a certain way, making some wonder whether there could be more efficiency. The value may have gotten split, dropping valuations for some and raising valuations for others. I don’t think this is a market meltdown à la the dot-com crash, I think it’s about a company or two getting revalued due to other companies being undervalued. The belief in the technology, the model and potential growth has not diminished and the belief that this is going to become a mainstream consumer and enterprise product is as strong as it could be. The technology can go mainstream faster, cheaper and maybe, even better.
The “cheaper” aspect has been playing out majorly in the minds of market players. Maybe, the sadist in us was glad to see something capable of challenging OpenAI, because we like to hate Sam Altman. He’s been such a marketeer and when he came to India, he had the guts and audacity to remark that it was hopeless to compete with his company. So, all of us were rejoicing that the DeepSeek team was able to accomplish that at a fraction of the cost.
But the reality is that OpenAI might be the BigTech of AI and may even build that. There’s an allegation that maybe, DeepSeek was, also, built on OpenAI’s model, so, maybe, whatever would be built would have some remnants of OpenAI. Is that a fair assumption?
Keith: It’s ironic that OpenAI might be complaining that someone is copying them when it has actually treated everyone else’s copyright as non-existent. Comparing AI models, ChatGPT is really good when it’s given instructions, whereas DeepSeek is a much more fluid thinker. The way they operate is quite different with different results to the same question, so I don’t think there’s a direct comparison. I think different AI models are being used for different purposes.
I think some of the statements that Altman made are going to come back to bite him. If you make statements like that, these sorts of things are going to happen. If $10 million can be deployed to build DeepSeek, how many DeepSeeks can be built in India with that kind of budget? It’s about smart engineers getting together and doing something special.
When the DeepSeek phenomenon happened, what were internal conversations like? Has the investing lens changed since then in terms of business models you’d want to look at now?
Bhandari: We mostly make investments in companies based out of India and many of the pitches that come to us are largely using a certain capability within the AI spectrum to go after a particular use case. So far, we haven’t looked at a generic AI company, so, the conversation now is that if it’s going to get cheaper and better, then any SaaS company automating a business process or making it more efficient should have a strong AI component, when it’s coming to us. And that AI component doesn’t mean a voice-activated chatbot or similar functions; there have to be ingenious uses to actually help companies.
Also Read: ChatGPT rolls out a new grumpy AI voice mode “Monday” for faster responses- All details
AI, for the sake of AI, isn’t going to help anybody. It has to be clear how the enterprise becomes more efficient or compliant or is able to get more business with the help of the AI model. No matter what’s being sold, it’s the balance sheet and the amount of money being returned to investors that are going to matter. If one can introduce technology that would improve the cost of operations or help a business grow, that’s what’s going to matter. So, we’re going to look at SaaS, tech and enterprise B2B companies more carefully, but, also, focus on how they’re utilizing AI smartly.
While AI may be a horizontal technology, there may be a lot of AI-washing taking place both by investors and entrepreneurs. Entrepreneurs may now add “.ai” to the suffix of their company names to market and position themselves as AI companies. There may even be a mushrooming of VC funds claiming to be AI-focused funds, because they may think it’s music to the ears of their LPs. So, how could someone see through the AI-washing and be confident that a business model is robust, sustainable and building for the long term?
Bhandari: Consider the flip side of this, where all this innovation is taking place and there aren’t enough VCs calling themselves “tech investors” or “DeepTech investors”. There will be those who create the conversation rather than just be in it and there are those who are thinking about what the conversation ought to be in the future to predict and prepare for that. The role of investors in the tech industry and similar spaces is to figure out which ventures deserve patient capital.
Keith: When we talk about SaaS, we might traditionally refer to “Software-as-a-Service”, but it could, soon, mean “Service-as-a-Software” and that’s how this goes from horizontal to vertical. The ability to monetize AI hasn’t been all there, so one has to focus on how services are being delivered. There have to be lots of vertical deep plays, because if everyone tries to be horizontal, there won’t be enough space.
So, what is the ideal product-market fit for an AI startup? How does one distinguish between a good AI company and a not-so-great one?
Keith: The perfect PMF is when one doesn’t even think or care about whether there’s AI in the company. The customers just enjoy the benefits of the company, be it Netflix recommendations or directions from Google Maps, without ever thinking about the underlying technology.
Post Open-AI, if one were to build a solid formidable Internet business, does it need to have AI?
Bhandari: Every business will need AI. If they’re not using it, they will, most likely, cease to be businesses.
Keith: AI is at the core of what we do. If I look at how my team operates, close to 20% of our code is written by AI. We think we can get to north of 50% and it may cap out at around 75%. That means I can grow my topline and my R&D development capacity without increasing headcount. I don’t think enterprises are using AI to the capacity that they should.
Look at some of the previous waves in digital innovation. There was the rise of e-commerce, then crypto, then the Metaverse and now, it’s AI. Nobody even talks about the Metaverse as much as they used to. Even Facebook changed its name to “Meta” and now, no one knows where the Metaverse boom went. So, how does one discern whether AI is a multi-generational opportunity? At the same time, tech visionaries seem to be warning people about the dangers of AI as if a new animal has been unleashed. How can one tell that this isn’t going to be a fad? Where’s the conviction coming from?
Bhandari: I think it’s about need. AI is impacting people’s livelihoods more than their lifestyle. Previous waves seemed to be lifestyle-oriented. Interestingly, the Metaverse was able to transform a lot of home-based businesses and allow them to sell their products on Instagram or Facebook or WhatsApp. So, it’s about getting companies to experiment with business models and reach more customers not previously anticipated. The Jevons paradox is going to play out much faster and bigger than one can imagine.
Also Read: Data for training stored overseas, copyright law doesn’t apply: OpenAI
Keith: AI’s been around for decades. It’s in the public consciousness right now, but people have been using Alexa or Siri long before this. So, AI is already everywhere, it’s just that ChatGPT has brought it into the mainstream. AI is going to absolutely be part of our lives going forward in the form of multiple use cases and in ways we haven’t thought about yet.
Bhandari: I think there might even be a future where machines might propose laws. I don’t see the day too far when societies could be governed digitally. One could see that as a negative in the form of machines controlling humans or one could make productive use of this to create guardrails for people.
When it comes to the creative aspect of AI impacting the media industry, I think we get what we deserve. Ultimately, AI is going to be such a horizontal technology that’s cheap and so abundantly available, that it would be commoditised. So, are we ready for that?
And what do you think are the common mistakes founders should avoid as they build AI businesses? Are we going to see an AI-only company becoming big or will we see companies getting big on the back of using AI?
Bhandari: If it’s not improving productivity, profits, operations, compliance or not making the planet greener, it’s unlikely that it’s going to get funded.
Keith: One should never assume what they’re building is unique. Assume you’re in second place and you’ll work real hard to get to first place. If you think you don’t have competitors, you probably don’t have PMF.
Shrija Agrawal is a business journalist. The views expressed are personal